Re: [rfa] ALL_OBJFILE_MSYMBOLS

[David Carlton <carlton@math.stanford.edu>]

On 31 Jan 2003 15:01:24 -0500, Jim Blandy <jimb@redhat.com> said:

> I think I'd prefer the patch below.  Could you try it out and see if
> it works as well?

Yes, it also fixes the bug I'm seeing.

> This was messier than I had expected.

Yah.

> (Which is a great example of one reason C macros suck and Lisp
> macros don't --- in case that's an emotionally charged issue for
> anyone else out there like it is for me.  :) )

I thought for a while about good responses to this, but I think I'll
just leave it alone for now. :-)

> But it would be better anyway for an empty minsym table to have a
> single, consistent representation.

That was my first reaction, too.  But while I was thinking about how
to do that, I noticed that ALL_MSYMBOLS already dealt with the
possibility of a NULL entry, so I just went with that out of laziness.

Like you, trusting the count sounds like a good idea.  And, even if
you go with a terminating entry, having it be a fake symbol instead
of, say, a NULL pointer is pretty weird, too.  But making either of
those changes sounds like too much (fallible) work for too little
benefit.

> So here's a patch which simply ensures that every objfile's minsym
> table has a terminating entry, and makes some appropriate accompanying
> changes.  The tests are still running, but I haven't noticed any
> regressions yet.

Seems sensible to me.  I didn't do a full testsuite run, but I did
apply it to my branch and try to trigger the bug I'd been seeing
there, and your patch does protect against the bug.

David Carlton
carlton@math.stanford.edu